#!/bin/sh
# graceful-drain.sh — preStop hook for IGW Pod hc container
# POSIX/ash compatible (alpine). Each step prints [N] description.
# Best-effort: individual curl failures are logged but do not abort the script.

set -u

HC_URL="${HC_URL:-http://127.0.0.1:18180}"
ENVOY_ADMIN="${ENVOY_ADMIN:-http://127.0.0.1:15000}"
DRAIN_TIMEOUT="${DRAIN_TIMEOUT:-120}"
POLL_INTERVAL="${POLL_INTERVAL:-2}"
LB_BUFFER="${LB_BUFFER:-10}"

START_TS=$(date +%s)

log() { printf "ts=%s %s\n" "$(date -u +%Y-%m-%dT%H:%M:%SZ)" "$*"; }

# curl wrapper — logs failures, never exits
do_curl() {
    out=$(curl -sf -m 5 "$@" 2>&1)
    rc=$?
    if [ $rc -ne 0 ]; then
        log "level=warn msg=curl_failed args=$* rc=$rc output=${out}"
    fi
    printf '%s' "$out"
    return $rc
}

# ---------------------------------------------------------------------------
# [1] Transition hc → DRAINING
# ---------------------------------------------------------------------------
log "[1] transitioning hc to DRAINING"
do_curl -X POST "${HC_URL}/drain/start" || true

# ---------------------------------------------------------------------------
# [2] Signal Envoy to drain listeners (graceful, don't exit)
# ---------------------------------------------------------------------------
log "[2] posting drain_listeners to Envoy admin"
do_curl -X POST "${ENVOY_ADMIN}/drain_listeners?graceful&skip_exit" || true

# ---------------------------------------------------------------------------
# [3] Poll active request counts until 0 or timeout
# ---------------------------------------------------------------------------
log "[3] polling Envoy stats for active requests (timeout=${DRAIN_TIMEOUT}s)"
DEADLINE=$((START_TS + DRAIN_TIMEOUT))

while true; do
    NOW=$(date +%s)

    # Fetch stats; on curl failure treat as non-zero so we keep polling
    STATS=$(curl -sf -m 5 \
        "${ENVOY_ADMIN}/stats?filter=downstream_rq_active%7Cupstream_rq_active" 2>/dev/null || true)

    # Sum all matching gauge values using awk
    ACTIVE=$(printf '%s\n' "$STATS" | awk -F': ' '
        /downstream_rq_active|upstream_rq_active/ { sum += $2 }
        END { print (sum+0) }
    ')

    log "level=info msg=drain_poll active_requests=${ACTIVE}"

    if [ "${ACTIVE}" -eq 0 ] 2>/dev/null; then
        log "level=info msg=all_requests_drained"
        break
    fi

    if [ "$NOW" -ge "$DEADLINE" ]; then
        log "level=warn msg=drain_timeout_exceeded elapsed=$((NOW - START_TS))s"
        break
    fi

    sleep "${POLL_INTERVAL}"
done

# ---------------------------------------------------------------------------
# [4] Flip /health_check.html → 503 (remove from HAProxy backend pool)
# ---------------------------------------------------------------------------
log "[4] transitioning hc to DRAINED_WAIT_LB (/health_check.html → 503)"
do_curl -X POST "${HC_URL}/drain/lb-fail" || true

# ---------------------------------------------------------------------------
# [5] Wait for HAProxy to detect the health failure and remove backend
# ---------------------------------------------------------------------------
log "[5] sleeping LB_BUFFER=${LB_BUFFER}s for HAProxy to drain backend"
sleep "${LB_BUFFER}"

# ---------------------------------------------------------------------------
# [6] Flip /health → 503 (K8s removes endpoint from Service)
# ---------------------------------------------------------------------------
log "[6] transitioning hc to TERMINATING (/health → 503)"
do_curl -X POST "${HC_URL}/disable-readiness" || true

# ---------------------------------------------------------------------------
# [7] Done
# ---------------------------------------------------------------------------
ELAPSED=$(( $(date +%s) - START_TS ))
log "[done] preStop completed in ${ELAPSED}s"